Design of an Android-Based Sitting Posture Detection Application Using Deep Learning

Jhonshen Lim; Octara Pribadi; Andy

doi:10.59934/jaiea.v5i3.2341

Authors

Jhonshen Lim STMIK Time
Octara Pribadi STMIK TIME
Andy STMIK TIME

DOI:

https://doi.org/10.59934/jaiea.v5i3.2341

Keywords:

Sitting Posture, Deep Learning, Multi-Layer Perceptron, MediaPipe BlazePose, TensorFlow Lite, Android, Pose Estimation

Abstract

Prolonged poor sitting posture is a major cause of musculoskeletal disorders including lower back pain and spinal abnormalities. This study designs and implements PosturApp, a deep learning-based Android application for real-time sitting posture detection using Kotlin. A Multi-Layer Perceptron (MLP) model was trained on 3,526 keypoint datasets sourced from the Kaggle public dataset (Posture Recognition) and direct image capture using an Android front camera, extracting 66 coordinate values from 33 body landmarks via MediaPipe BlazePose. The model was converted to TensorFlow Lite (TFLite) format at approximately 78 KB for on-device inference without internet connectivity. Evaluation results show an accuracy of 97.81% with precision 0.99, recall 0.99, and F1-Score 0.98. The application provides real-time visual feedback through interface color changes and corrective notifications, along with a gallery-based classification feature. Functional testing across eight posture scenarios yielded entirely correct results with confidence values ranging from 59% to 99%.

Downloads

Download data is not yet available.

References

Ienaga, N., Takahata, S., Terayama, K., & others. (2022). Development and verification of postural control assessment using deep-learning-based pose estimators. Occupational Therapy International, 2022, 1–11. https://doi.org/10.1155/2022/6952999

Wu, J., Li, W., Chen, Y., & others. (2024). A human pose estimation method based on the BlazePose model. Journal of Modern Educational Research, 2024, 1–10. https://doi.org/10.53964/jmer.2024021

Swain, D., Satapathy, S. K., Acharya, B., & others. (2022). Deep learning models for yoga pose monitoring. Algorithms, 15(11), 403. https://doi.org/10.3390/a15110403

Krishna, R. (2023). Real time human body posture analysis using deep learning. International Journal for Science Technology and Engineering, 4(3), 1–6. https://doi.org/10.22214/ijraset.2023.52099

Ranjan, R., Ahmedt-Aristizabal, D., Armin, M. A., & others. (2024). Developing normative gait cycle parameters for clinical analysis using human pose estimation. arXiv preprint, 2411.13716. https://doi.org/10.48550/arxiv.2411.13716

Chung, J. L., Ong, L. Y., & Leow, M. C. (2022). Comparative analysis of skeleton-based human pose estimation. Future Internet, 14(12), 380. https://doi.org/10.3390/fi14120380

Torun, C., & Karaci, A. (2024). Comparison of 2D human pose estimation models based on deep learning. Journal of Sustainable Engineering Applications and Technological Developments, 4(2), 1–10.

Haghighat, M. (2023). Introduction to computer vision with convolutional neural networks. Edge AI and Vision Alliance.

Yauri, R., & Espino, R. (2023). Edge device for movement pattern classification using neural network algorithms. Indonesian Journal of Electrical Engineering and Computer Science, 30(1), 229–236.

Romindo, R., Barus, O. P., Pangaribuan, J. J., & others. (2022). Implementation of support vector machine algorithm for ballet pose classification. Building of Informatics, Technology and Science, 4(3), 1–7.

Object-Based Hybrid Deep Learning Technique for Recognition of Sequential Actions. (2023). IEEE Access, 1–12. https://doi.org/10.1109/access.2023.3291395

Chiddarwar, G. G., Ranjane, A., Chindhe, M., & others. (2020). AI-based yoga pose estimation for Android application. International Journal of Innovative Science and Research Technology, 5(10), 30–36.